Generative AI and copyright

If a work generated with the use of generative AI is creative and original, consider it to be protected and follow the normal copyright steps for clearing, crediting, using etc.

Check the terms of the tool you use to make sure they specify that the work you create is yours to use.
Cite and credit your source - always.
Credit yourself and note that an AI was used it you create a text or an image using generative AI.
Be transparent about the process through which you arrived at a particular result. Always mark Al-generated material as Al-generated. Don't use generative Al to fool others.
Any content you produce is your responsibility. Take responsibility for every text, image, sound or other product that you create with Al. That means making sure factual statements are correct, that sources are properly credited, that unlawful and unethical output is deleted, and that discriminatory or plagiarized content is avoided.

- just same as with any other tool or service you use when creating works of literature or art.

laws on copyright and generative ai

There are no specific copyright exceptions or permissions in the law regarding the use of generative AI (apart from a few possibilities for text and data mining, TDM), and so far, there are no decisions by courts in Europe that we can use as a guide.

Output generated by/with generative AI

For now, the legal consensus is that for output generated by or with the help of generative AI we simply follow the same copyright rules as we would with works created with any other service or tool.

Data used to train generative AI

Much of the data used to train the big commercial AI systems is almost certainly from copyrighted works, and has often been used without permission or payment. This is difficult to prove as many of the systems are "black boxes".

Until the AI-companies and rightsholders sort this issue out through licensing or in the courts, the DTU Legal Office does not consider using the various generative AI tools an infringement on the copyright of the materials that might have used in training the AI.

Plagiarism is still a thing

However, output can still be considered plagiarism if it too closely resembles an existing original work.

Anything you have created yourself (your own work)
Anything in the public domain
Anything given a CC-0 license by the rightsholder

Not much else, unless:

You get permission from the rightsholder
You are very specifically doing TDM as a part of a scientific research project, because of the TDM exception (§11c) in the Danish copyright law

What about Open Access articles etc.?

Use only the CC-BY licensed articles, and
Cite everything, and make a reference list of all uploaded materials, to satisfy the attribution requirement

This applies to all the tools/services, including Microsoft Co-pilot with DTU login.

No, AI generated images are not exempt from copyright. You should follow the normal copyright rules and procedures when using an image, regardless of the tools used to make it: get permission, cite the source.

If you create an image using generative AI, remember to credit yourself and check the terms of the tool you use, as you should with any tool or service you use.

If you create an image using generative AI, remember to credit yourself. The starting point is that you are the creator and rightsholder, however, we advise you to check the terms of the AI service you use, just as you would with any tool or service you use.

You should make a note that AI was used in the creation of the image.

You credit yourself by adding your name and the AI used. You could also choose to add the title of the image (if you have given it a title) and the date of creation and version of the AI, but these are optional.

Save your prompts

From a copyright perspective, it is extremely important that you save your prompts when working with generative AI.

Your prompts are the only proof that you are the creator of the work.

An example of referencing an AI generated image

If you wanted to use a picture of a cat in a hat for your slides, you might go to DALL-E via the DTU approved Microsoft Co-pilot, and use the prompt "present me with an illustration or picture of a 'scientist cat in a hat'". This would generate several images*, and you can choose one.

For example this one:

AI generated image of a “scientist cat in a hat.” by J. Ekstrøm on DALL-E. Used under a CC-BY license.

As you can see, the reference is short and easy to add. We have also added a note of the license the creator has added to the image - you would not need this for your own image.

A more complete reference could look like this (here using APA referencing style):

Ekstrøm, J. (2024). An AI generated image of a “scientist cat in a hat.” Dall E 3 via Microsoft Copilot. Retrieved from https://copilot.microsoft.com/.

Please note: for any image (whether generated using AI or not) to be protected by copyright, it must be creative and original. A short prompt like the one used here would probably be so unoriginal that the image does not reach the "threshold of originality".

However - protected or not, the image should always be cited and referenced correctly. And images generated by others should be considered theirs for referencing and permission purposes.

*And yes, it is interesting that the cat in the image isn't in the hat, as prompted. Which goes to show that to get an image you actually want to use, you often need several iterations and prompts, which would mean that you would be creative and original.

Contact

The copyright group at DTU ophavsret@dtu.dk

originality

For any work of art or literature (whether generated using AI or not) to be protected by copyright, it must be creative and original.

However, there are no firm guidelines on when a work is creative or original enough to reach the "threshold of originality".

When working with generative AI

a short prompt will most likely not be considered to be enough for your work to be protected by copyright.
you will most likely need several iterations and creative decisions to get a result you want to use, in which case your work should reach the threshold of originality.

text and data mining

Danish Copyright Law, as well as the EU DSM Directive, contains an exception that allows universities to do text and data mining (TDM) on materials we have legal access to.

TDM is defined as “any automated analytical technique aimed at analysing text and data in digital form in order to generate information which includes but is not limited to patterns, trends and correlations”.

This exception is only valid if the materials are from publishers in the EU.

Contact copyright@dtu.dk or elibrary@dtu.dk for more information.